- Knet Knet is a traditional machine learning framework with a proper performance. You can use it to build up your ML applications without much worries.
- ForwardDiff It is a forward mode automatic differentiation for Julia, with a state of the art performance. Check the paper for details.
- NiLang NiLang is a differential reversible eDSL, any program written in this eDSL can be differentiated efficiently. Check the paper for details.
- Flux Flux is the most popular machine learning library in Julia. Check the paper for detail.
- ReverseDiff It is a reverse mode autodiff tool that can differentiate a general Julia program with a proper performance.
- Zygote Zygote is an easy to use autodiff framework based on source code transformation, it can differentiate a general Julia program.
- Nabla
- Yota
None of the above machine learning packages can be as stable as popular python packages like TensorFlow and Pytorch. Need some try and errors.